fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
   4
   5   This program can be distributed under the terms of the GNU GPL.
   6   See the file COPYING.
   7 */
   8
   9 #include "fuse_i.h"
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/init.h>
  16 #include <linux/module.h>
  17 #include <linux/moduleparam.h>
  18 #include <linux/fs_context.h>
  19 #include <linux/fs_parser.h>
  20 #include <linux/statfs.h>
  21 #include <linux/random.h>
  22 #include <linux/sched.h>
  23 #include <linux/exportfs.h>
  24 #include <linux/posix_acl.h>
  25 #include <linux/pid_namespace.h>
  26
  27 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  28 MODULE_DESCRIPTION("Filesystem in Userspace");
  29 MODULE_LICENSE("GPL");
  30
  31 static struct kmem_cache *fuse_inode_cachep;
  32 struct list_head fuse_conn_list;
  33 DEFINE_MUTEX(fuse_mutex);
  34
  35 static int set_global_limit(const char *val, const struct kernel_param *kp);
  36
  37 unsigned max_user_bgreq;
  38 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  39                   &max_user_bgreq, 0644);
  40 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  41 MODULE_PARM_DESC(max_user_bgreq,
  42  "Global limit for the maximum number of backgrounded requests an "
  43  "unprivileged user can set");
  44
  45 unsigned max_user_congthresh;
  46 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  47                   &max_user_congthresh, 0644);
  48 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  49 MODULE_PARM_DESC(max_user_congthresh,
  50  "Global limit for the maximum congestion threshold an "
  51  "unprivileged user can set");
  52
  53 #define FUSE_SUPER_MAGIC 0x65735546
  54
  55 #define FUSE_DEFAULT_BLKSIZE 512
  56
  57 /** Maximum number of outstanding background requests */
  58 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  59
  60 /** Congestion starts at 75% of maximum */
  61 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  62
  63 #ifdef CONFIG_BLOCK
  64 static struct file_system_type fuseblk_fs_type;
  65 #endif
  66
  67 struct fuse_forget_link *fuse_alloc_forget(void)
  68 {
  69         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
  70 }
  71
  72 static struct inode *fuse_alloc_inode(struct super_block *sb)
  73 {
  74         struct fuse_inode *fi;
  75
  76         fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  77         if (!fi)
  78                 return NULL;
  79
  80         fi->i_time = 0;
  81         fi->inval_mask = 0;
  82         fi->nodeid = 0;
  83         fi->nlookup = 0;
  84         fi->attr_version = 0;
  85         fi->orig_ino = 0;
  86         fi->state = 0;
  87         mutex_init(&fi->mutex);
  88         spin_lock_init(&fi->lock);
  89         fi->forget = fuse_alloc_forget();
  90         if (!fi->forget) {
  91                 kmem_cache_free(fuse_inode_cachep, fi);
  92                 return NULL;
  93         }
  94
  95         return &fi->inode;
  96 }
  97
  98 static void fuse_free_inode(struct inode *inode)
  99 {
 100         struct fuse_inode *fi = get_fuse_inode(inode);
 101
 102         mutex_destroy(&fi->mutex);
 103         kfree(fi->forget);
 104         kmem_cache_free(fuse_inode_cachep, fi);
 105 }
 106
 107 static void fuse_evict_inode(struct inode *inode)
 108 {
 109         struct fuse_inode *fi = get_fuse_inode(inode);
 110
 111         truncate_inode_pages_final(&inode->i_data);
 112         clear_inode(inode);
 113         if (inode->i_sb->s_flags & SB_ACTIVE) {
 114                 struct fuse_conn *fc = get_fuse_conn(inode);
 115                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 116                 fi->forget = NULL;
 117         }
 118         if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
 119                 WARN_ON(!list_empty(&fi->write_files));
 120                 WARN_ON(!list_empty(&fi->queued_writes));
 121         }
 122 }
 123
 124 static int fuse_reconfigure(struct fs_context *fc)
 125 {
 126         struct super_block *sb = fc->root->d_sb;
 127
 128         sync_filesystem(sb);
 129         if (fc->sb_flags & SB_MANDLOCK)
 130                 return -EINVAL;
 131
 132         return 0;
 133 }
 134
 135 /*
 136  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 137  * so that it will fit.
 138  */
 139 static ino_t fuse_squash_ino(u64 ino64)
 140 {
 141         ino_t ino = (ino_t) ino64;
 142         if (sizeof(ino_t) < sizeof(u64))
 143                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 144         return ino;
 145 }
 146
 147 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 148                                    u64 attr_valid)
 149 {
 150         struct fuse_conn *fc = get_fuse_conn(inode);
 151         struct fuse_inode *fi = get_fuse_inode(inode);
 152
 153         lockdep_assert_held(&fi->lock);
 154
 155         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 156         fi->i_time = attr_valid;
 157         WRITE_ONCE(fi->inval_mask, 0);
 158
 159         inode->i_ino     = fuse_squash_ino(attr->ino);
 160         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 161         set_nlink(inode, attr->nlink);
 162         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
 163         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
 164         inode->i_blocks  = attr->blocks;
 165         inode->i_atime.tv_sec   = attr->atime;
 166         inode->i_atime.tv_nsec  = attr->atimensec;
 167         /* mtime from server may be stale due to local buffered write */
 168         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
 169                 inode->i_mtime.tv_sec   = attr->mtime;
 170                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 171                 inode->i_ctime.tv_sec   = attr->ctime;
 172                 inode->i_ctime.tv_nsec  = attr->ctimensec;
 173         }
 174
 175         if (attr->blksize != 0)
 176                 inode->i_blkbits = ilog2(attr->blksize);
 177         else
 178                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 179
 180         /*
 181          * Don't set the sticky bit in i_mode, unless we want the VFS
 182          * to check permissions.  This prevents failures due to the
 183          * check in may_delete().
 184          */
 185         fi->orig_i_mode = inode->i_mode;
 186         if (!fc->default_permissions)
 187                 inode->i_mode &= ~S_ISVTX;
 188
 189         fi->orig_ino = attr->ino;
 190 }
 191
 192 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 193                             u64 attr_valid, u64 attr_version)
 194 {
 195         struct fuse_conn *fc = get_fuse_conn(inode);
 196         struct fuse_inode *fi = get_fuse_inode(inode);
 197         bool is_wb = fc->writeback_cache;
 198         loff_t oldsize;
 199         struct timespec64 old_mtime;
 200
 201         spin_lock(&fi->lock);
 202         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 203             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 204                 spin_unlock(&fi->lock);
 205                 return;
 206         }
 207
 208         old_mtime = inode->i_mtime;
 209         fuse_change_attributes_common(inode, attr, attr_valid);
 210
 211         oldsize = inode->i_size;
 212         /*
 213          * In case of writeback_cache enabled, the cached writes beyond EOF
 214          * extend local i_size without keeping userspace server in sync. So,
 215          * attr->size coming from server can be stale. We cannot trust it.
 216          */
 217         if (!is_wb || !S_ISREG(inode->i_mode))
 218                 i_size_write(inode, attr->size);
 219         spin_unlock(&fi->lock);
 220
 221         if (!is_wb && S_ISREG(inode->i_mode)) {
 222                 bool inval = false;
 223
 224                 if (oldsize != attr->size) {
 225                         truncate_pagecache(inode, attr->size);
 226                         if (!fc->explicit_inval_data)
 227                                 inval = true;
 228                 } else if (fc->auto_inval_data) {
 229                         struct timespec64 new_mtime = {
 230                                 .tv_sec = attr->mtime,
 231                                 .tv_nsec = attr->mtimensec,
 232                         };
 233
 234                         /*
 235                          * Auto inval mode also checks and invalidates if mtime
 236                          * has changed.
 237                          */
 238                         if (!timespec64_equal(&old_mtime, &new_mtime))
 239                                 inval = true;
 240                 }
 241
 242                 if (inval)
 243                         invalidate_inode_pages2(inode->i_mapping);
 244         }
 245 }
 246
 247 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 248 {
 249         inode->i_mode = attr->mode & S_IFMT;
 250         inode->i_size = attr->size;
 251         inode->i_mtime.tv_sec  = attr->mtime;
 252         inode->i_mtime.tv_nsec = attr->mtimensec;
 253         inode->i_ctime.tv_sec  = attr->ctime;
 254         inode->i_ctime.tv_nsec = attr->ctimensec;
 255         if (S_ISREG(inode->i_mode)) {
 256                 fuse_init_common(inode);
 257                 fuse_init_file_inode(inode);
 258         } else if (S_ISDIR(inode->i_mode))
 259                 fuse_init_dir(inode);
 260         else if (S_ISLNK(inode->i_mode))
 261                 fuse_init_symlink(inode);
 262         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 263                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 264                 fuse_init_common(inode);
 265                 init_special_inode(inode, inode->i_mode,
 266                                    new_decode_dev(attr->rdev));
 267         } else
 268                 BUG();
 269 }
 270
 271 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 272 {
 273         u64 nodeid = *(u64 *) _nodeidp;
 274         if (get_node_id(inode) == nodeid)
 275                 return 1;
 276         else
 277                 return 0;
 278 }
 279
 280 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 281 {
 282         u64 nodeid = *(u64 *) _nodeidp;
 283         get_fuse_inode(inode)->nodeid = nodeid;
 284         return 0;
 285 }
 286
 287 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 288                         int generation, struct fuse_attr *attr,
 289                         u64 attr_valid, u64 attr_version)
 290 {
 291         struct inode *inode;
 292         struct fuse_inode *fi;
 293         struct fuse_conn *fc = get_fuse_conn_super(sb);
 294
 295  retry:
 296         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 297         if (!inode)
 298                 return NULL;
 299
 300         if ((inode->i_state & I_NEW)) {
 301                 inode->i_flags |= S_NOATIME;
 302                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 303                         inode->i_flags |= S_NOCMTIME;
 304                 inode->i_generation = generation;
 305                 fuse_init_inode(inode, attr);
 306                 unlock_new_inode(inode);
 307         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 308                 /* Inode has changed type, any I/O on the old should fail */
 309                 fuse_make_bad(inode);
 310                 iput(inode);
 311                 goto retry;
 312         }
 313
 314         fi = get_fuse_inode(inode);
 315         spin_lock(&fi->lock);
 316         fi->nlookup++;
 317         spin_unlock(&fi->lock);
 318         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 319
 320         return inode;
 321 }
 322
 323 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 324                              loff_t offset, loff_t len)
 325 {
 326         struct fuse_conn *fc = get_fuse_conn_super(sb);
 327         struct fuse_inode *fi;
 328         struct inode *inode;
 329         pgoff_t pg_start;
 330         pgoff_t pg_end;
 331
 332         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 333         if (!inode)
 334                 return -ENOENT;
 335
 336         fi = get_fuse_inode(inode);
 337         spin_lock(&fi->lock);
 338         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 339         spin_unlock(&fi->lock);
 340
 341         fuse_invalidate_attr(inode);
 342         forget_all_cached_acls(inode);
 343         if (offset >= 0) {
 344                 pg_start = offset >> PAGE_SHIFT;
 345                 if (len <= 0)
 346                         pg_end = -1;
 347                 else
 348                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 349                 invalidate_inode_pages2_range(inode->i_mapping,
 350                                               pg_start, pg_end);
 351         }
 352         iput(inode);
 353         return 0;
 354 }
 355
 356 bool fuse_lock_inode(struct inode *inode)
 357 {
 358         bool locked = false;
 359
 360         if (!get_fuse_conn(inode)->parallel_dirops) {
 361                 mutex_lock(&get_fuse_inode(inode)->mutex);
 362                 locked = true;
 363         }
 364
 365         return locked;
 366 }
 367
 368 void fuse_unlock_inode(struct inode *inode, bool locked)
 369 {
 370         if (locked)
 371                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 372 }
 373
 374 static void fuse_umount_begin(struct super_block *sb)
 375 {
 376         struct fuse_conn *fc = get_fuse_conn_super(sb);
 377
 378         if (!fc->no_force_umount)
 379                 fuse_abort_conn(fc);
 380 }
 381
 382 static void fuse_send_destroy(struct fuse_conn *fc)
 383 {
 384         if (fc->conn_init) {
 385                 FUSE_ARGS(args);
 386
 387                 args.opcode = FUSE_DESTROY;
 388                 args.force = true;
 389                 args.nocreds = true;
 390                 fuse_simple_request(fc, &args);
 391         }
 392 }
 393
 394 static void fuse_put_super(struct super_block *sb)
 395 {
 396         struct fuse_conn *fc = get_fuse_conn_super(sb);
 397
 398         mutex_lock(&fuse_mutex);
 399         list_del(&fc->entry);
 400         fuse_ctl_remove_conn(fc);
 401         mutex_unlock(&fuse_mutex);
 402
 403         fuse_conn_put(fc);
 404 }
 405
 406 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 407 {
 408         stbuf->f_type    = FUSE_SUPER_MAGIC;
 409         stbuf->f_bsize   = attr->bsize;
 410         stbuf->f_frsize  = attr->frsize;
 411         stbuf->f_blocks  = attr->blocks;
 412         stbuf->f_bfree   = attr->bfree;
 413         stbuf->f_bavail  = attr->bavail;
 414         stbuf->f_files   = attr->files;
 415         stbuf->f_ffree   = attr->ffree;
 416         stbuf->f_namelen = attr->namelen;
 417         /* fsid is left zero */
 418 }
 419
 420 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 421 {
 422         struct super_block *sb = dentry->d_sb;
 423         struct fuse_conn *fc = get_fuse_conn_super(sb);
 424         FUSE_ARGS(args);
 425         struct fuse_statfs_out outarg;
 426         int err;
 427
 428         if (!fuse_allow_current_process(fc)) {
 429                 buf->f_type = FUSE_SUPER_MAGIC;
 430                 return 0;
 431         }
 432
 433         memset(&outarg, 0, sizeof(outarg));
 434         args.in_numargs = 0;
 435         args.opcode = FUSE_STATFS;
 436         args.nodeid = get_node_id(d_inode(dentry));
 437         args.out_numargs = 1;
 438         args.out_args[0].size = sizeof(outarg);
 439         args.out_args[0].value = &outarg;
 440         err = fuse_simple_request(fc, &args);
 441         if (!err)
 442                 convert_fuse_statfs(buf, &outarg.st);
 443         return err;
 444 }
 445
 446 enum {
 447         OPT_SOURCE,
 448         OPT_SUBTYPE,
 449         OPT_FD,
 450         OPT_ROOTMODE,
 451         OPT_USER_ID,
 452         OPT_GROUP_ID,
 453         OPT_DEFAULT_PERMISSIONS,
 454         OPT_ALLOW_OTHER,
 455         OPT_MAX_READ,
 456         OPT_BLKSIZE,
 457         OPT_ERR
 458 };
 459
 460 static const struct fs_parameter_spec fuse_param_specs[] = {
 461         fsparam_string  ("source",              OPT_SOURCE),
 462         fsparam_u32     ("fd",                  OPT_FD),
 463         fsparam_u32oct  ("rootmode",            OPT_ROOTMODE),
 464         fsparam_u32     ("user_id",             OPT_USER_ID),
 465         fsparam_u32     ("group_id",            OPT_GROUP_ID),
 466         fsparam_flag    ("default_permissions", OPT_DEFAULT_PERMISSIONS),
 467         fsparam_flag    ("allow_other",         OPT_ALLOW_OTHER),
 468         fsparam_u32     ("max_read",            OPT_MAX_READ),
 469         fsparam_u32     ("blksize",             OPT_BLKSIZE),
 470         fsparam_string  ("subtype",             OPT_SUBTYPE),
 471         {}
 472 };
 473
 474 static const struct fs_parameter_description fuse_fs_parameters = {
 475         .name           = "fuse",
 476         .specs          = fuse_param_specs,
 477 };
 478
 479 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
 480 {
 481         struct fs_parse_result result;
 482         struct fuse_fs_context *ctx = fc->fs_private;
 483         int opt;
 484
 485         /*
 486          * Ignore options coming from mount(MS_REMOUNT) for backward
 487          * compatibility.
 488          */
 489         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE)
 490                 return 0;
 491
 492         opt = fs_parse(fc, &fuse_fs_parameters, param, &result);
 493         if (opt < 0)
 494                 return opt;
 495
 496         switch (opt) {
 497         case OPT_SOURCE:
 498                 if (fc->source)
 499                         return invalf(fc, "fuse: Multiple sources specified");
 500                 fc->source = param->string;
 501                 param->string = NULL;
 502                 break;
 503
 504         case OPT_SUBTYPE:
 505                 if (ctx->subtype)
 506                         return invalf(fc, "fuse: Multiple subtypes specified");
 507                 ctx->subtype = param->string;
 508                 param->string = NULL;
 509                 return 0;
 510
 511         case OPT_FD:
 512                 ctx->fd = result.uint_32;
 513                 ctx->fd_present = 1;
 514                 break;
 515
 516         case OPT_ROOTMODE:
 517                 if (!fuse_valid_type(result.uint_32))
 518                         return invalf(fc, "fuse: Invalid rootmode");
 519                 ctx->rootmode = result.uint_32;
 520                 ctx->rootmode_present = 1;
 521                 break;
 522
 523         case OPT_USER_ID:
 524                 ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
 525                 if (!uid_valid(ctx->user_id))
 526                         return invalf(fc, "fuse: Invalid user_id");
 527                 ctx->user_id_present = 1;
 528                 break;
 529
 530         case OPT_GROUP_ID:
 531                 ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
 532                 if (!gid_valid(ctx->group_id))
 533                         return invalf(fc, "fuse: Invalid group_id");
 534                 ctx->group_id_present = 1;
 535                 break;
 536
 537         case OPT_DEFAULT_PERMISSIONS:
 538                 ctx->default_permissions = 1;
 539                 break;
 540
 541         case OPT_ALLOW_OTHER:
 542                 ctx->allow_other = 1;
 543                 break;
 544
 545         case OPT_MAX_READ:
 546                 ctx->max_read = result.uint_32;
 547                 break;
 548
 549         case OPT_BLKSIZE:
 550                 if (!ctx->is_bdev)
 551                         return invalf(fc, "fuse: blksize only supported for fuseblk");
 552                 ctx->blksize = result.uint_32;
 553                 break;
 554
 555         default:
 556                 return -EINVAL;
 557         }
 558
 559         return 0;
 560 }
 561
 562 static void fuse_free_fc(struct fs_context *fc)
 563 {
 564         struct fuse_fs_context *ctx = fc->fs_private;
 565
 566         if (ctx) {
 567                 kfree(ctx->subtype);
 568                 kfree(ctx);
 569         }
 570 }
 571
 572 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 573 {
 574         struct super_block *sb = root->d_sb;
 575         struct fuse_conn *fc = get_fuse_conn_super(sb);
 576
 577         if (fc->no_mount_options)
 578                 return 0;
 579
 580         seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
 581         seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
 582         if (fc->default_permissions)
 583                 seq_puts(m, ",default_permissions");
 584         if (fc->allow_other)
 585                 seq_puts(m, ",allow_other");
 586         if (fc->max_read != ~0)
 587                 seq_printf(m, ",max_read=%u", fc->max_read);
 588         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 589                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 590         return 0;
 591 }
 592
 593 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
 594                              const struct fuse_iqueue_ops *ops,
 595                              void *priv)
 596 {
 597         memset(fiq, 0, sizeof(struct fuse_iqueue));
 598         spin_lock_init(&fiq->lock);
 599         init_waitqueue_head(&fiq->waitq);
 600         INIT_LIST_HEAD(&fiq->pending);
 601         INIT_LIST_HEAD(&fiq->interrupts);
 602         fiq->forget_list_tail = &fiq->forget_list_head;
 603         fiq->connected = 1;
 604         fiq->ops = ops;
 605         fiq->priv = priv;
 606 }
 607
 608 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 609 {
 610         unsigned int i;
 611
 612         spin_lock_init(&fpq->lock);
 613         for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
 614                 INIT_LIST_HEAD(&fpq->processing[i]);
 615         INIT_LIST_HEAD(&fpq->io);
 616         fpq->connected = 1;
 617 }
 618
 619 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns,
 620                     const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
 621 {
 622         memset(fc, 0, sizeof(*fc));
 623         spin_lock_init(&fc->lock);
 624         spin_lock_init(&fc->bg_lock);
 625         init_rwsem(&fc->killsb);
 626         refcount_set(&fc->count, 1);
 627         atomic_set(&fc->dev_count, 1);
 628         init_waitqueue_head(&fc->blocked_waitq);
 629         fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
 630         INIT_LIST_HEAD(&fc->bg_queue);
 631         INIT_LIST_HEAD(&fc->entry);
 632         INIT_LIST_HEAD(&fc->devices);
 633         atomic_set(&fc->num_waiting, 0);
 634         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 635         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 636         atomic64_set(&fc->khctr, 0);
 637         fc->polled_files = RB_ROOT;
 638         fc->blocked = 0;
 639         fc->initialized = 0;
 640         fc->connected = 1;
 641         atomic64_set(&fc->attr_version, 1);
 642         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 643         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 644         fc->user_ns = get_user_ns(user_ns);
 645         fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
 646 }
 647 EXPORT_SYMBOL_GPL(fuse_conn_init);
 648
 649 void fuse_conn_put(struct fuse_conn *fc)
 650 {
 651         if (refcount_dec_and_test(&fc->count)) {
 652                 struct fuse_iqueue *fiq = &fc->iq;
 653
 654                 if (fiq->ops->release)
 655                         fiq->ops->release(fiq);
 656                 put_pid_ns(fc->pid_ns);
 657                 put_user_ns(fc->user_ns);
 658                 fc->release(fc);
 659         }
 660 }
 661 EXPORT_SYMBOL_GPL(fuse_conn_put);
 662
 663 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 664 {
 665         refcount_inc(&fc->count);
 666         return fc;
 667 }
 668 EXPORT_SYMBOL_GPL(fuse_conn_get);
 669
 670 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 671 {
 672         struct fuse_attr attr;
 673         memset(&attr, 0, sizeof(attr));
 674
 675         attr.mode = mode;
 676         attr.ino = FUSE_ROOT_ID;
 677         attr.nlink = 1;
 678         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 679 }
 680
 681 struct fuse_inode_handle {
 682         u64 nodeid;
 683         u32 generation;
 684 };
 685
 686 static struct dentry *fuse_get_dentry(struct super_block *sb,
 687                                       struct fuse_inode_handle *handle)
 688 {
 689         struct fuse_conn *fc = get_fuse_conn_super(sb);
 690         struct inode *inode;
 691         struct dentry *entry;
 692         int err = -ESTALE;
 693
 694         if (handle->nodeid == 0)
 695                 goto out_err;
 696
 697         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 698         if (!inode) {
 699                 struct fuse_entry_out outarg;
 700                 const struct qstr name = QSTR_INIT(".", 1);
 701
 702                 if (!fc->export_support)
 703                         goto out_err;
 704
 705                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 706                                        &inode);
 707                 if (err && err != -ENOENT)
 708                         goto out_err;
 709                 if (err || !inode) {
 710                         err = -ESTALE;
 711                         goto out_err;
 712                 }
 713                 err = -EIO;
 714                 if (get_node_id(inode) != handle->nodeid)
 715                         goto out_iput;
 716         }
 717         err = -ESTALE;
 718         if (inode->i_generation != handle->generation)
 719                 goto out_iput;
 720
 721         entry = d_obtain_alias(inode);
 722         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 723                 fuse_invalidate_entry_cache(entry);
 724
 725         return entry;
 726
 727  out_iput:
 728         iput(inode);
 729  out_err:
 730         return ERR_PTR(err);
 731 }
 732
 733 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 734                            struct inode *parent)
 735 {
 736         int len = parent ? 6 : 3;
 737         u64 nodeid;
 738         u32 generation;
 739
 740         if (*max_len < len) {
 741                 *max_len = len;
 742                 return  FILEID_INVALID;
 743         }
 744
 745         nodeid = get_fuse_inode(inode)->nodeid;
 746         generation = inode->i_generation;
 747
 748         fh[0] = (u32)(nodeid >> 32);
 749         fh[1] = (u32)(nodeid & 0xffffffff);
 750         fh[2] = generation;
 751
 752         if (parent) {
 753                 nodeid = get_fuse_inode(parent)->nodeid;
 754                 generation = parent->i_generation;
 755
 756                 fh[3] = (u32)(nodeid >> 32);
 757                 fh[4] = (u32)(nodeid & 0xffffffff);
 758                 fh[5] = generation;
 759         }
 760
 761         *max_len = len;
 762         return parent ? 0x82 : 0x81;
 763 }
 764
 765 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 766                 struct fid *fid, int fh_len, int fh_type)
 767 {
 768         struct fuse_inode_handle handle;
 769
 770         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 771                 return NULL;
 772
 773         handle.nodeid = (u64) fid->raw[0] << 32;
 774         handle.nodeid |= (u64) fid->raw[1];
 775         handle.generation = fid->raw[2];
 776         return fuse_get_dentry(sb, &handle);
 777 }
 778
 779 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 780                 struct fid *fid, int fh_len, int fh_type)
 781 {
 782         struct fuse_inode_handle parent;
 783
 784         if (fh_type != 0x82 || fh_len < 6)
 785                 return NULL;
 786
 787         parent.nodeid = (u64) fid->raw[3] << 32;
 788         parent.nodeid |= (u64) fid->raw[4];
 789         parent.generation = fid->raw[5];
 790         return fuse_get_dentry(sb, &parent);
 791 }
 792
 793 static struct dentry *fuse_get_parent(struct dentry *child)
 794 {
 795         struct inode *child_inode = d_inode(child);
 796         struct fuse_conn *fc = get_fuse_conn(child_inode);
 797         struct inode *inode;
 798         struct dentry *parent;
 799         struct fuse_entry_out outarg;
 800         const struct qstr name = QSTR_INIT("..", 2);
 801         int err;
 802
 803         if (!fc->export_support)
 804                 return ERR_PTR(-ESTALE);
 805
 806         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 807                                &name, &outarg, &inode);
 808         if (err) {
 809                 if (err == -ENOENT)
 810                         return ERR_PTR(-ESTALE);
 811                 return ERR_PTR(err);
 812         }
 813
 814         parent = d_obtain_alias(inode);
 815         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 816                 fuse_invalidate_entry_cache(parent);
 817
 818         return parent;
 819 }
 820
 821 static const struct export_operations fuse_export_operations = {
 822         .fh_to_dentry   = fuse_fh_to_dentry,
 823         .fh_to_parent   = fuse_fh_to_parent,
 824         .encode_fh      = fuse_encode_fh,
 825         .get_parent     = fuse_get_parent,
 826 };
 827
 828 static const struct super_operations fuse_super_operations = {
 829         .alloc_inode    = fuse_alloc_inode,
 830         .free_inode     = fuse_free_inode,
 831         .evict_inode    = fuse_evict_inode,
 832         .write_inode    = fuse_write_inode,
 833         .drop_inode     = generic_delete_inode,
 834         .put_super      = fuse_put_super,
 835         .umount_begin   = fuse_umount_begin,
 836         .statfs         = fuse_statfs,
 837         .show_options   = fuse_show_options,
 838 };
 839
 840 static void sanitize_global_limit(unsigned *limit)
 841 {
 842         /*
 843          * The default maximum number of async requests is calculated to consume
 844          * 1/2^13 of the total memory, assuming 392 bytes per request.
 845          */
 846         if (*limit == 0)
 847                 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
 848
 849         if (*limit >= 1 << 16)
 850                 *limit = (1 << 16) - 1;
 851 }
 852
 853 static int set_global_limit(const char *val, const struct kernel_param *kp)
 854 {
 855         int rv;
 856
 857         rv = param_set_uint(val, kp);
 858         if (rv)
 859                 return rv;
 860
 861         sanitize_global_limit((unsigned *)kp->arg);
 862
 863         return 0;
 864 }
 865
 866 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 867 {
 868         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 869
 870         if (arg->minor < 13)
 871                 return;
 872
 873         sanitize_global_limit(&max_user_bgreq);
 874         sanitize_global_limit(&max_user_congthresh);
 875
 876         spin_lock(&fc->bg_lock);
 877         if (arg->max_background) {
 878                 fc->max_background = arg->max_background;
 879
 880                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 881                         fc->max_background = max_user_bgreq;
 882         }
 883         if (arg->congestion_threshold) {
 884                 fc->congestion_threshold = arg->congestion_threshold;
 885
 886                 if (!cap_sys_admin &&
 887                     fc->congestion_threshold > max_user_congthresh)
 888                         fc->congestion_threshold = max_user_congthresh;
 889         }
 890         spin_unlock(&fc->bg_lock);
 891 }
 892
 893 struct fuse_init_args {
 894         struct fuse_args args;
 895         struct fuse_init_in in;
 896         struct fuse_init_out out;
 897 };
 898
 899 static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
 900                                int error)
 901 {
 902         struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
 903         struct fuse_init_out *arg = &ia->out;
 904
 905         if (error || arg->major != FUSE_KERNEL_VERSION)
 906                 fc->conn_error = 1;
 907         else {
 908                 unsigned long ra_pages;
 909
 910                 process_init_limits(fc, arg);
 911
 912                 if (arg->minor >= 6) {
 913                         ra_pages = arg->max_readahead / PAGE_SIZE;
 914                         if (arg->flags & FUSE_ASYNC_READ)
 915                                 fc->async_read = 1;
 916                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 917                                 fc->no_lock = 1;
 918                         if (arg->minor >= 17) {
 919                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 920                                         fc->no_flock = 1;
 921                         } else {
 922                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 923                                         fc->no_flock = 1;
 924                         }
 925                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 926                                 fc->atomic_o_trunc = 1;
 927                         if (arg->minor >= 9) {
 928                                 /* LOOKUP has dependency on proto version */
 929                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 930                                         fc->export_support = 1;
 931                         }
 932                         if (arg->flags & FUSE_BIG_WRITES)
 933                                 fc->big_writes = 1;
 934                         if (arg->flags & FUSE_DONT_MASK)
 935                                 fc->dont_mask = 1;
 936                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 937                                 fc->auto_inval_data = 1;
 938                         else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
 939                                 fc->explicit_inval_data = 1;
 940                         if (arg->flags & FUSE_DO_READDIRPLUS) {
 941                                 fc->do_readdirplus = 1;
 942                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
 943                                         fc->readdirplus_auto = 1;
 944                         }
 945                         if (arg->flags & FUSE_ASYNC_DIO)
 946                                 fc->async_dio = 1;
 947                         if (arg->flags & FUSE_WRITEBACK_CACHE)
 948                                 fc->writeback_cache = 1;
 949                         if (arg->flags & FUSE_PARALLEL_DIROPS)
 950                                 fc->parallel_dirops = 1;
 951                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
 952                                 fc->handle_killpriv = 1;
 953                         if (arg->time_gran && arg->time_gran <= 1000000000)
 954                                 fc->sb->s_time_gran = arg->time_gran;
 955                         if ((arg->flags & FUSE_POSIX_ACL)) {
 956                                 fc->default_permissions = 1;
 957                                 fc->posix_acl = 1;
 958                                 fc->sb->s_xattr = fuse_acl_xattr_handlers;
 959                         }
 960                         if (arg->flags & FUSE_CACHE_SYMLINKS)
 961                                 fc->cache_symlinks = 1;
 962                         if (arg->flags & FUSE_ABORT_ERROR)
 963                                 fc->abort_err = 1;
 964                         if (arg->flags & FUSE_MAX_PAGES) {
 965                                 fc->max_pages =
 966                                         min_t(unsigned int, FUSE_MAX_MAX_PAGES,
 967                                         max_t(unsigned int, arg->max_pages, 1));
 968                         }
 969                 } else {
 970                         ra_pages = fc->max_read / PAGE_SIZE;
 971                         fc->no_lock = 1;
 972                         fc->no_flock = 1;
 973                 }
 974
 975                 fc->sb->s_bdi->ra_pages =
 976                                 min(fc->sb->s_bdi->ra_pages, ra_pages);
 977                 fc->minor = arg->minor;
 978                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 979                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 980                 fc->conn_init = 1;
 981         }
 982         kfree(ia);
 983
 984         fuse_set_initialized(fc);
 985         wake_up_all(&fc->blocked_waitq);
 986 }
 987
 988 void fuse_send_init(struct fuse_conn *fc)
 989 {
 990         struct fuse_init_args *ia;
 991
 992         ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
 993
 994         ia->in.major = FUSE_KERNEL_VERSION;
 995         ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
 996         ia->in.max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
 997         ia->in.flags |=
 998                 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 999                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1000                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1001                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1002                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1003                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1004                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1005                 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1006                 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
1007         ia->args.opcode = FUSE_INIT;
1008         ia->args.in_numargs = 1;
1009         ia->args.in_args[0].size = sizeof(ia->in);
1010         ia->args.in_args[0].value = &ia->in;
1011         ia->args.out_numargs = 1;
1012         /* Variable length argument used for backward compatibility
1013            with interface version < 7.5.  Rest of init_out is zeroed
1014            by do_get_request(), so a short reply is not a problem */
1015         ia->args.out_argvar = 1;
1016         ia->args.out_args[0].size = sizeof(ia->out);
1017         ia->args.out_args[0].value = &ia->out;
1018         ia->args.force = true;
1019         ia->args.nocreds = true;
1020         ia->args.end = process_init_reply;
1021
1022         if (fuse_simple_background(fc, &ia->args, GFP_KERNEL) != 0)
1023                 process_init_reply(fc, &ia->args, -ENOTCONN);
1024 }
1025 EXPORT_SYMBOL_GPL(fuse_send_init);
1026
1027 void fuse_free_conn(struct fuse_conn *fc)
1028 {
1029         WARN_ON(!list_empty(&fc->devices));
1030         kfree_rcu(fc, rcu);
1031 }
1032 EXPORT_SYMBOL_GPL(fuse_free_conn);
1033
1034 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1035 {
1036         int err;
1037         char *suffix = "";
1038
1039         if (sb->s_bdev) {
1040                 suffix = "-fuseblk";
1041                 /*
1042                  * sb->s_bdi points to blkdev's bdi however we want to redirect
1043                  * it to our private bdi...
1044                  */
1045                 bdi_put(sb->s_bdi);
1046                 sb->s_bdi = &noop_backing_dev_info;
1047         }
1048         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1049                                    MINOR(fc->dev), suffix);
1050         if (err)
1051                 return err;
1052
1053         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1054         /* fuse does it's own writeback accounting */
1055         sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1056
1057         /*
1058          * For a single fuse filesystem use max 1% of dirty +
1059          * writeback threshold.
1060          *
1061          * This gives about 1M of write buffer for memory maps on a
1062          * machine with 1G and 10% dirty_ratio, which should be more
1063          * than enough.
1064          *
1065          * Privileged users can raise it by writing to
1066          *
1067          *    /sys/class/bdi/<bdi>/max_ratio
1068          */
1069         bdi_set_max_ratio(sb->s_bdi, 1);
1070
1071         return 0;
1072 }
1073
1074 struct fuse_dev *fuse_dev_alloc(void)
1075 {
1076         struct fuse_dev *fud;
1077         struct list_head *pq;
1078
1079         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1080         if (!fud)
1081                 return NULL;
1082
1083         pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1084         if (!pq) {
1085                 kfree(fud);
1086                 return NULL;
1087         }
1088
1089         fud->pq.processing = pq;
1090         fuse_pqueue_init(&fud->pq);
1091
1092         return fud;
1093 }
1094 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1095
1096 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1097 {
1098         fud->fc = fuse_conn_get(fc);
1099         spin_lock(&fc->lock);
1100         list_add_tail(&fud->entry, &fc->devices);
1101         spin_unlock(&fc->lock);
1102 }
1103 EXPORT_SYMBOL_GPL(fuse_dev_install);
1104
1105 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1106 {
1107         struct fuse_dev *fud;
1108
1109         fud = fuse_dev_alloc();
1110         if (!fud)
1111                 return NULL;
1112
1113         fuse_dev_install(fud, fc);
1114         return fud;
1115 }
1116 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1117
1118 void fuse_dev_free(struct fuse_dev *fud)
1119 {
1120         struct fuse_conn *fc = fud->fc;
1121
1122         if (fc) {
1123                 spin_lock(&fc->lock);
1124                 list_del(&fud->entry);
1125                 spin_unlock(&fc->lock);
1126
1127                 fuse_conn_put(fc);
1128         }
1129         kfree(fud->pq.processing);
1130         kfree(fud);
1131 }
1132 EXPORT_SYMBOL_GPL(fuse_dev_free);
1133
1134 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1135 {
1136         struct fuse_dev *fud;
1137         struct fuse_conn *fc = get_fuse_conn_super(sb);
1138         struct inode *root;
1139         struct dentry *root_dentry;
1140         int err;
1141
1142         err = -EINVAL;
1143         if (sb->s_flags & SB_MANDLOCK)
1144                 goto err;
1145
1146         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1147
1148         if (ctx->is_bdev) {
1149 #ifdef CONFIG_BLOCK
1150                 err = -EINVAL;
1151                 if (!sb_set_blocksize(sb, ctx->blksize))
1152                         goto err;
1153 #endif
1154         } else {
1155                 sb->s_blocksize = PAGE_SIZE;
1156                 sb->s_blocksize_bits = PAGE_SHIFT;
1157         }
1158
1159         sb->s_subtype = ctx->subtype;
1160         ctx->subtype = NULL;
1161         sb->s_magic = FUSE_SUPER_MAGIC;
1162         sb->s_op = &fuse_super_operations;
1163         sb->s_xattr = fuse_xattr_handlers;
1164         sb->s_maxbytes = MAX_LFS_FILESIZE;
1165         sb->s_time_gran = 1;
1166         sb->s_export_op = &fuse_export_operations;
1167         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1168         if (sb->s_user_ns != &init_user_ns)
1169                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1170
1171         /*
1172          * If we are not in the initial user namespace posix
1173          * acls must be translated.
1174          */
1175         if (sb->s_user_ns != &init_user_ns)
1176                 sb->s_xattr = fuse_no_acl_xattr_handlers;
1177
1178         fud = fuse_dev_alloc_install(fc);
1179         if (!fud)
1180                 goto err;
1181
1182         fc->dev = sb->s_dev;
1183         fc->sb = sb;
1184         err = fuse_bdi_init(fc, sb);
1185         if (err)
1186                 goto err_dev_free;
1187
1188         /* Handle umasking inside the fuse code */
1189         if (sb->s_flags & SB_POSIXACL)
1190                 fc->dont_mask = 1;
1191         sb->s_flags |= SB_POSIXACL;
1192
1193         fc->default_permissions = ctx->default_permissions;
1194         fc->allow_other = ctx->allow_other;
1195         fc->user_id = ctx->user_id;
1196         fc->group_id = ctx->group_id;
1197         fc->max_read = max_t(unsigned, 4096, ctx->max_read);
1198         fc->destroy = ctx->destroy;
1199         fc->no_control = ctx->no_control;
1200         fc->no_force_umount = ctx->no_force_umount;
1201         fc->no_mount_options = ctx->no_mount_options;
1202
1203         err = -ENOMEM;
1204         root = fuse_get_root_inode(sb, ctx->rootmode);
1205         sb->s_d_op = &fuse_root_dentry_operations;
1206         root_dentry = d_make_root(root);
1207         if (!root_dentry)
1208                 goto err_dev_free;
1209         /* Root dentry doesn't have .d_revalidate */
1210         sb->s_d_op = &fuse_dentry_operations;
1211
1212         mutex_lock(&fuse_mutex);
1213         err = -EINVAL;
1214         if (*ctx->fudptr)
1215                 goto err_unlock;
1216
1217         err = fuse_ctl_add_conn(fc);
1218         if (err)
1219                 goto err_unlock;
1220
1221         list_add_tail(&fc->entry, &fuse_conn_list);
1222         sb->s_root = root_dentry;
1223         *ctx->fudptr = fud;
1224         mutex_unlock(&fuse_mutex);
1225         return 0;
1226
1227  err_unlock:
1228         mutex_unlock(&fuse_mutex);
1229         dput(root_dentry);
1230  err_dev_free:
1231         fuse_dev_free(fud);
1232  err:
1233         return err;
1234 }
1235 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1236
1237 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1238 {
1239         struct fuse_fs_context *ctx = fsc->fs_private;
1240         struct file *file;
1241         int err;
1242         struct fuse_conn *fc;
1243
1244         err = -EINVAL;
1245         file = fget(ctx->fd);
1246         if (!file)
1247                 goto err;
1248
1249         /*
1250          * Require mount to happen from the same user namespace which
1251          * opened /dev/fuse to prevent potential attacks.
1252          */
1253         if ((file->f_op != &fuse_dev_operations) ||
1254             (file->f_cred->user_ns != sb->s_user_ns))
1255                 goto err_fput;
1256         ctx->fudptr = &file->private_data;
1257
1258         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1259         err = -ENOMEM;
1260         if (!fc)
1261                 goto err_fput;
1262
1263         fuse_conn_init(fc, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1264         fc->release = fuse_free_conn;
1265         sb->s_fs_info = fc;
1266
1267         err = fuse_fill_super_common(sb, ctx);
1268         if (err)
1269                 goto err_put_conn;
1270         /*
1271          * atomic_dec_and_test() in fput() provides the necessary
1272          * memory barrier for file->private_data to be visible on all
1273          * CPUs after this
1274          */
1275         fput(file);
1276         fuse_send_init(get_fuse_conn_super(sb));
1277         return 0;
1278
1279  err_put_conn:
1280         fuse_conn_put(fc);
1281         sb->s_fs_info = NULL;
1282  err_fput:
1283         fput(file);
1284  err:
1285         return err;
1286 }
1287
1288 static int fuse_get_tree(struct fs_context *fc)
1289 {
1290         struct fuse_fs_context *ctx = fc->fs_private;
1291
1292         if (!ctx->fd_present || !ctx->rootmode_present ||
1293             !ctx->user_id_present || !ctx->group_id_present)
1294                 return -EINVAL;
1295
1296 #ifdef CONFIG_BLOCK
1297         if (ctx->is_bdev)
1298                 return get_tree_bdev(fc, fuse_fill_super);
1299 #endif
1300
1301         return get_tree_nodev(fc, fuse_fill_super);
1302 }
1303
1304 static const struct fs_context_operations fuse_context_ops = {
1305         .free           = fuse_free_fc,
1306         .parse_param    = fuse_parse_param,
1307         .reconfigure    = fuse_reconfigure,
1308         .get_tree       = fuse_get_tree,
1309 };
1310
1311 /*
1312  * Set up the filesystem mount context.
1313  */
1314 static int fuse_init_fs_context(struct fs_context *fc)
1315 {
1316         struct fuse_fs_context *ctx;
1317
1318         ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1319         if (!ctx)
1320                 return -ENOMEM;
1321
1322         ctx->max_read = ~0;
1323         ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1324
1325 #ifdef CONFIG_BLOCK
1326         if (fc->fs_type == &fuseblk_fs_type) {
1327                 ctx->is_bdev = true;
1328                 ctx->destroy = true;
1329         }
1330 #endif
1331
1332         fc->fs_private = ctx;
1333         fc->ops = &fuse_context_ops;
1334         return 0;
1335 }
1336
1337 static void fuse_sb_destroy(struct super_block *sb)
1338 {
1339         struct fuse_conn *fc = get_fuse_conn_super(sb);
1340
1341         if (fc) {
1342                 if (fc->destroy)
1343                         fuse_send_destroy(fc);
1344
1345                 fuse_abort_conn(fc);
1346                 fuse_wait_aborted(fc);
1347
1348                 down_write(&fc->killsb);
1349                 fc->sb = NULL;
1350                 up_write(&fc->killsb);
1351         }
1352 }
1353
1354 void fuse_kill_sb_anon(struct super_block *sb)
1355 {
1356         fuse_sb_destroy(sb);
1357         kill_anon_super(sb);
1358 }
1359 EXPORT_SYMBOL_GPL(fuse_kill_sb_anon);
1360
1361 static struct file_system_type fuse_fs_type = {
1362         .owner          = THIS_MODULE,
1363         .name           = "fuse",
1364         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1365         .init_fs_context = fuse_init_fs_context,
1366         .parameters     = &fuse_fs_parameters,
1367         .kill_sb        = fuse_kill_sb_anon,
1368 };
1369 MODULE_ALIAS_FS("fuse");
1370
1371 #ifdef CONFIG_BLOCK
1372 static void fuse_kill_sb_blk(struct super_block *sb)
1373 {
1374         fuse_sb_destroy(sb);
1375         kill_block_super(sb);
1376 }
1377
1378 static struct file_system_type fuseblk_fs_type = {
1379         .owner          = THIS_MODULE,
1380         .name           = "fuseblk",
1381         .init_fs_context = fuse_init_fs_context,
1382         .parameters     = &fuse_fs_parameters,
1383         .kill_sb        = fuse_kill_sb_blk,
1384         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1385 };
1386 MODULE_ALIAS_FS("fuseblk");
1387
1388 static inline int register_fuseblk(void)
1389 {
1390         return register_filesystem(&fuseblk_fs_type);
1391 }
1392
1393 static inline void unregister_fuseblk(void)
1394 {
1395         unregister_filesystem(&fuseblk_fs_type);
1396 }
1397 #else
1398 static inline int register_fuseblk(void)
1399 {
1400         return 0;
1401 }
1402
1403 static inline void unregister_fuseblk(void)
1404 {
1405 }
1406 #endif
1407
1408 static void fuse_inode_init_once(void *foo)
1409 {
1410         struct inode *inode = foo;
1411
1412         inode_init_once(inode);
1413 }
1414
1415 static int __init fuse_fs_init(void)
1416 {
1417         int err;
1418
1419         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1420                         sizeof(struct fuse_inode), 0,
1421                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1422                         fuse_inode_init_once);
1423         err = -ENOMEM;
1424         if (!fuse_inode_cachep)
1425                 goto out;
1426
1427         err = register_fuseblk();
1428         if (err)
1429                 goto out2;
1430
1431         err = register_filesystem(&fuse_fs_type);
1432         if (err)
1433                 goto out3;
1434
1435         return 0;
1436
1437  out3:
1438         unregister_fuseblk();
1439  out2:
1440         kmem_cache_destroy(fuse_inode_cachep);
1441  out:
1442         return err;
1443 }
1444
1445 static void fuse_fs_cleanup(void)
1446 {
1447         unregister_filesystem(&fuse_fs_type);
1448         unregister_fuseblk();
1449
1450         /*
1451          * Make sure all delayed rcu free inodes are flushed before we
1452          * destroy cache.
1453          */
1454         rcu_barrier();
1455         kmem_cache_destroy(fuse_inode_cachep);
1456 }
1457
1458 static struct kobject *fuse_kobj;
1459
1460 static int fuse_sysfs_init(void)
1461 {
1462         int err;
1463
1464         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1465         if (!fuse_kobj) {
1466                 err = -ENOMEM;
1467                 goto out_err;
1468         }
1469
1470         err = sysfs_create_mount_point(fuse_kobj, "connections");
1471         if (err)
1472                 goto out_fuse_unregister;
1473
1474         return 0;
1475
1476  out_fuse_unregister:
1477         kobject_put(fuse_kobj);
1478  out_err:
1479         return err;
1480 }
1481
1482 static void fuse_sysfs_cleanup(void)
1483 {
1484         sysfs_remove_mount_point(fuse_kobj, "connections");
1485         kobject_put(fuse_kobj);
1486 }
1487
1488 static int __init fuse_init(void)
1489 {
1490         int res;
1491
1492         pr_info("init (API version %i.%i)\n",
1493                 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1494
1495         INIT_LIST_HEAD(&fuse_conn_list);
1496         res = fuse_fs_init();
1497         if (res)
1498                 goto err;
1499
1500         res = fuse_dev_init();
1501         if (res)
1502                 goto err_fs_cleanup;
1503
1504         res = fuse_sysfs_init();
1505         if (res)
1506                 goto err_dev_cleanup;
1507
1508         res = fuse_ctl_init();
1509         if (res)
1510                 goto err_sysfs_cleanup;
1511
1512         sanitize_global_limit(&max_user_bgreq);
1513         sanitize_global_limit(&max_user_congthresh);
1514
1515         return 0;
1516
1517  err_sysfs_cleanup:
1518         fuse_sysfs_cleanup();
1519  err_dev_cleanup:
1520         fuse_dev_cleanup();
1521  err_fs_cleanup:
1522         fuse_fs_cleanup();
1523  err:
1524         return res;
1525 }
1526
1527 static void __exit fuse_exit(void)
1528 {
1529         pr_debug("exit\n");
1530
1531         fuse_ctl_cleanup();
1532         fuse_sysfs_cleanup();
1533         fuse_fs_cleanup();
1534         fuse_dev_cleanup();
1535 }
1536
1537 module_init(fuse_init);
1538 module_exit(fuse_exit);